Accurate risk prediction and early diagnosis are crucial for the early intervention of Alzheimer's disease (AD). Current prediction models usually have limited power in capturing the complex interplays between the heterogeneous inputs or lack biological explainability required for clinical adoption and new diagnosis biomarkers discovery. Inspired by pioneering works on biologically informed network and multi-modal learning, we presented an Explainable Imaging-Genetic Neural Network (EIGNN), integrating genetic and neuroimaging data to generate accurate, robust, and explainable AD risk prediction. The EIGNN model features a biologically-informed architecture, incorporating a multi-GWAS SNP selection strategy, enhanced explainable neural network design, and a modal attention mechanism. Genetic variants were hierarchically mapped to their target genes and biological pathways, and further integrated with neuroimaging features. We demonstrated that the EIGNN model outperformed existing methods and exhibited improved robustness, explainability, and reproducibility. Finally, by applying a novel biologically informed multi-modal feature interaction map, we prioritized a set of AD risk genes and biological pathways, and explored the intricate interactions between the risk genes and brain regions implicated in AD risk.

Brain imaging genomics is an emerging interdisciplinary field, where integrated analysis of multimodal medical image-derived phenotypes (IDPs) and multi-omics data, bridging the gap between macroscopic brain phenotypes and their cellular and molecular characteristics. This approach aims to better interpret the genetic architecture and molecular mechanisms associated with brain structure, function and clinical outcomes. More recently, the availability of large-scale imaging and multi-omics datasets from the human brain has afforded the opportunity to the discovering of common genetic variants contributing to the structural and functional IDPs of the human brain. By integrative analyses with functional multi-omics data from the human brain, a set of critical genes, functional genomic regions and neuronal cell types have been identified as significantly associated with brain IDPs. Here, we review the recent advances in the methods and applications of multi-omics integration in brain imaging analysis. We highlight the importance of functional genomic datasets in understanding the biological functions of the identified genes and cell types that are associated with brain IDPs. Moreover, we summarize well-known neuroimaging genetics datasets and discuss challenges and future directions in this field.

Positron emission tomography (PET) with fluorodeoxyglucose (FDG) or florbetapir (AV45) has been proved effective in the diagnosis of Alzheimer's disease. However, the expensive and radioactive nature of PET has limited its application. Here, employing multi-layer perceptron mixer architecture, we present a deep learning model, namely 3-dimensional multi-task multi-layer perceptron mixer, for simultaneously predicting the standardized uptake value ratios (SUVRs) for FDG-PET and AV45-PET from the cheap and widely used structural magnetic resonance imaging data, and the model can be further used for Alzheimer's disease diagnosis based on embedding features derived from SUVR prediction. Experiment results demonstrate the high prediction accuracy of the proposed method for FDG/AV45-PET SUVRs, where we achieved Pearson's correlation coefficients of 0.66 and 0.61 respectively between the estimated and actual SUVR and the estimated SUVRs also show high sensitivity and distinct longitudinal patterns for different disease status. By taking into account PET embedding features, the proposed method outperforms other competing methods on five independent datasets in the diagnosis of Alzheimer's disease and discriminating between stable and progressive mild cognitive impairments, achieving the area under receiver operating characteristic curves of 0.968 and 0.776 respectively on ADNI dataset, and generalizes better to other external datasets. Moreover, the top-weighted patches extracted from the trained model involve important brain regions related to Alzheimer's disease, suggesting good biological interpretability of our proposed method."

Exploring multimorbidity relationships among diseases is of great importance for understanding their shared mechanisms, precise diagnosis and treatment. However, the landscape of multimorbidities is still far from complete due to the complex nature of multimorbidity. Although various types of biological data, such as biomolecules and clinical symptoms, have been used to identify multimorbidities, the population phenotype information (e.g. physical activity and diet) remains less explored for multimorbidity. Here, we present a graph convolutional network (GCN) model, named MorbidGCN, for multimorbidity prediction by integrating population phenotypes and disease network. Specifically, MorbidGCN treats the multimorbidity prediction as a missing link prediction problem in the disease network, where a novel feature selection method is embedded to select important phenotypes. Benchmarking results on two large-scale multimorbidity data sets, i.e. the UK Biobank (UKB) and Human Disease Network (HuDiNe) data sets, demonstrate that MorbidGCN outperforms other competitive methods. With MorbidGCN, 9742 and 14 010 novel multimorbidities are identified in the UKB and HuDiNe data sets, respectively. Moreover, we notice that the selected phenotypes that are generally differentially distributed between multimorbidity patients and single-disease patients can help interpret multimorbidities and show potential for prognosis of multimorbidities.

Advances in high-throughput experimental technologies promote the accumulation of vast number of biomedical data. Biomedical link prediction and single-cell RNA-sequencing (scRNA-seq) data imputation are two essential tasks in biomedical data analyses, which can facilitate various downstream studies and gain insights into the mechanisms of complex diseases. Both tasks can be transformed into matrix completion problems. For a variety of matrix completion tasks, matrix factorization has shown promising performance. However, the sparseness and high dimensionality of biomedical networks and scRNA-seq data have raised new challenges. To resolve these issues, various matrix factorization methods have emerged recently. In this paper, we present a comprehensive review on such matrix factorization methods and their usage in biomedical link prediction and scRNA-seq data imputation. Moreover, we select representative matrix factorization methods and conduct a systematic empirical comparison on 15 real data sets to evaluate their performance under different scenarios. By summarizing the experimental results, we provide general guidelines for selecting matrix factorization methods for different biomedical matrix completion tasks and point out some future directions to further improve the performance for biomedical link prediction and scRNA-seq data imputation.

MOTIVATION: Predicting potential links in biomedical bipartite networks can provide useful insights into the diagnosis and treatment of complex diseases and the discovery of novel drug targets. Computational methods have been proposed recently to predict potential links for various biomedical bipartite networks. However, existing methods are usually rely on the coverage of known links, which may encounter difficulties when dealing with new nodes without any known link information. RESULTS: In this study, we propose a new link prediction method, named graph regularized generalized matrix factorization (GRGMF), to identify potential links in biomedical bipartite networks. First, we formulate a generalized matrix factorization model to exploit the latent patterns behind observed links. In particular, it can take into account the neighborhood information of each node when learning the latent representation for each node, and the neighborhood information of each node can be learned adaptively. Second, we introduce two graph regularization terms to draw support from affinity information of each node derived from external databases to enhance the learning of latent representations. We conduct extensive experiments on six real datasets. Experiment results show that GRGMF can achieve competitive performance on all these datasets, which demonstrate the effectiveness of GRGMF in prediction potential links in biomedical bipartite networks. AVAILABILITY AND IMPLEMENTATION: The package is available at https://github.com/happyalfred2016/GRGMF. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.